The 'it just works' approach might be great with simple scenarios, but it's horrible for complex use cases where you need to have detailed control over the pieces that will run in production and - once they are properly setup -, you would anyway automate accordingly your specific usecase.

Hence my: "Additional hidden complexity doesn't seem to be a good thing in this case".

Beside that, I really have no idea about what they specifically mean with "Kubernetes-based PaaS". What is it? Abstracting kubernetes cluster setup? If this is the case, i remark that I need more fine-grained access on the internals rather than additional abstractions that internally do "god-knows-what".